Imaging's Real Payoff: Character Amount Recognition
Mark Arend · ABA banking journal · 1993
As if on cue, the major image technology vendors announced, in the closing days of 1992, pilot testing or actual production versions of the most elusive piece of check imaging: character amount recognition (CAR) for proof of deposit. The hardest part of imaging technology to develop, character - or courtesy - amount recognition holds the most allure for bankers of any imaging function, including image statements, because of the direct savings in personnel costs it provides. CAR-equipped check processing systems can automatically and encode the handwritten amount in the courtesy boxes of checks, deposit tickets, and other items, reducing the number of employees needed to read the amounts and encode the items manually. In theory, if a machine can accurately read the courtesy amount of 50% of the items it handles, then 50% fewer encoders are needed to handle the same amount of work. (Actual character read rates as announced by imaging vendors depend on the mixture of items being read. A high concentration of personal checks, where the courtesy amount box is in a standard location, usually yields a higher read rate. A large percentage of business checks and other items tends to lower the read rate in CAR testing and production.) There is rarely an exact correlation between the number of items read automatically and the number of manhours saved. But bankers testing CAR systems from IBM, Unisys, NCR, and Banctec Systems are optimistic about what the technology can do once it's fully implemented in their organizations. Passing the test. We have achieved the target read rate for the end of 1993 already, reports Norma Reed, assistant vice-president and product administrator for image technology at The Huntington Service Co., a subsidiary of $12 billion-assets Huntington Bancshares, Columbus, Ohio. The target read rate is the average number of items bank management wants the imaging equipment to read and encode by a given date. When Huntington began using a CAR module of InfoImage, a high-speed item processing system from Unisys, in November, the system accurately read 35% to 36% of the items, says Reed. On Thursday, March 4, the figure rose to 44%, we were very excited about, says Reed. Huntington has set a more aggressive target read rate of 50% during years two through five of the installation. and the bank have some work to do to get to that number, says Reed. The reasons CAR read rates aren't higher at Huntington and other banks testing the technology include: (1) the condition or of bank documents; (2) the variety of ways consumers write numerals; and (3) the variety of locations on business checks where the courtesy amount box is located. Regarding the last point, CAR systems must, in effect, either learn where to look for the numeric information or be programmed to find it, whether the amount is handwritten or printed by a computer. Huntington Bancshares is working on the first issue by image-enabling its various documents, that is, redesigning them to conform to standards that facilitate image processing. Deposit, savings, and loan-payment tickets obtained in the branches, for example, now include special shaded boxes, dollar signs, and other features that will improve image read rates. We're now at the point where there are more image-enabled documents throughout the organization than not, says Reed. This exercise also helps the [nonimage] environment, because it is an opportunity to standardize your forms so that even the encoding operators gain throughput benefits. Finding the amount. Unisys is addressing the issues of numeral readability and courtesy box locations by developing and implementing X-Y coordinates, which are algorithms that help the system find the amount boxes on business checks and other items. A built-in algorithm detects and reads personal check amounts, which generally conform to American National Standards Institute standards. …